Custom Object Detection for Fresh and Rotten Fruit Classification

Automated grading and processing of fruits is a complex task which requires the creation of effective fruit vision detection systems. This paper presents an introduction of Fruit detection and recognition system for grading apples, bananas and oranges though incorporation of deep learning methodologies. For this study, two separate databases of fruits including the Kaggle and Google databases were carefully prepared. The images were annotated with precision and augmented using the Roboflow web-based platform, categorizing each image into one of six classes: fresh apple, rotten apple, fresh orange, rotten orange, fresh banana, rotten banana. After data collection, the next step involved was data partitioning into training and validation datasets through a data splitting technique. Regarding the model training process, the YOLOv8 and YOLOv5 model was used. The evaluation that was conducted on the trained model was impressive, as the system returned a high value for accuracy, precision, recall, F1-score, and mAP. The proposed system not only can enhance fruit grading efficiency but also has a wide range of applications in other agricultural fields. Moreover, this study helps toward some significant concerns in agricultural industry, particularly in the area of fruit recognition. It aims to improve the automation the automation of post-harvest fruit quality selection, ensuring more efficient and sorting and grading processes. This extensive review is quite beneficial as it provides insights into the future direction on how to better improve automatic fruit picking technology for better efficiency and optimization.